CoppeliaSim forums

Dear coppelia,

We are developing a simulation scene with several LIDAR, 3D sensors and a quite high amount of objects with a lot of triangles.
Since we are going to get hardware to work on this, my question is:
Can V-REP take advantage of a workstation graphics card as Nvidia Quadro M2000 ou M4000? as does CAD software?
Or would we get better performance using a gforce 1060,e.g.?

We have already part of the scene developed, and we can see a huge slow down when we handle the vision sensors (2 sick s300 + realsense + 3d sensor ).

Best,
Tiago

Hello Tiago,

V-REP currently still uses old plain openGl instructions for rendering (that is camera and vision sensor renderings). Lidars are normally also based on vision sensors. So a good graphic card will benefit, but it is difficult to say which one would produce the fastest renderings. The best option you have is to create low-poly models. You can still attach to the low-poly shapes high-poly shapes, that are hidden, but used in calculations.

Cheers

Dear coppelia,

Thank you!
My biggest problem is not with renderings, I have reduced most of the shapes triangles count. I am sure we can still improve this part.
Handling the vision sensors (LIDAR as you pointed out and 3D depth) and retrieving the data are the things taking the biggest time to execute.

I have not yet tried it, but does handling the vision sensor and retrieving its data with extra threads is allowed in V-REP(using the same thread for handling a specific sensor and retrieving its data)?

Best,
Tiago

With sensors such as lidars, there are several things involved:

• Reading one or several vision sensor images (timing directly affected by the graphic card)
• Processing the read data (e.g. coordinate transformations). This is not handled by the graphic card. This can be handled by a script (super slow), or via a plugin (much faster, CPU based). If this is handled via a plugin, you can have several threads operate on that at the same time (user created threads, not V-REP threads). There is also a vision sensor filter component that does something similar: Extract coordinates from work image
• Sometime also data display in the scene

Cheers

Dear coppelia,

Thank you for detailing who does what in terms of processing and using the embedded filters of VREP.

We will try to follow your advice!

Best,
Tiago